Ioan Ionita wrote: > Linus doesn't plan on making any header useable by userspace.
Not entirely true; from what I was reading in the archives, Linus doesn't mind making the kernel headers usable by glibc. But he absolutely does not want random userspace programs using them, because they will never follow namespace rules. (They'll always be creating structs, constants, etc., with names that might conflict with stuff that the userspace program wants to use.) glibc can use them, but he doesn't want glibc copying them into /usr/include/bits or something like that. > I don't quite understand his whole point. His point seemed to be, "look at /usr/include/bits/fcntl.h". When the size of "struct flock" changes depending on whether the userspace program wants large-file support or not, that's really hard to do in the kernel headers. He does *not* want the kernel headers to cater to this kind of thing, and I agree with him completely on that. This kind of thing is for glibc to do, not the kernel. (That's because fcntl.h is glibc's interface to the rest of userspace. It needs to define that interface well enough that random userspace programs will almost all work. In the same vein, though, the Linux kernel needs to define its interface to all of userspace, glibc or anything that wants to bypass glibc -- and right now it doesn't do that all that well. Yes, the syscall interface is stable, but it isn't documented; the easiest documentation is probably a set of headers.)
signature.asc
Description: OpenPGP digital signature
-- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
